Author: Gian-Carlo Pascutto
Date: 11:26:52 03/12/02
Go up one level in this thread
On March 12, 2002 at 14:17:18, William H Rogers wrote: >You can do what I do and that is to find another program that is 100 to 200 >points stronger than yours is. Then play a bunch of games, recording every move. >Make changes to your eval then play a bunch more to see if there is any change. >I found that some changes in evals do not effect both colors the same. Sometimes >you will find that a certain change will make black play stronger and then >others will favor white. It depends upon ...... >Hell, I don't really know. >Bill Thanks for the suggestion. I remember Don Daily writing once that he tested his program by letting it play matches from fixed starting positions, and then comparing the moves played before and after the eval change until a different move is played. If the new move is better, the eval change is good. It's a workable system, but it still requires human judgement. I'd ideally want something that can be more or less automated. Playing testmatches doesn't cut it; too much variance in the results to test small changes and too slow too. I have a testing system for search changes that is more or less automated, fairly accurate and reasonably fast. Constructing something similar for eval changes would be great, but I don't know it it's even possible. -- GCP
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.